Abstract

MALDONADO SERRANO, Jorge Francisco  and  CADENA ZAMBRANO, Andrés Felipe. Large Language Models (LLMs) as Word Calculators: A Conceptual Clarification. Prax. filos. [online]. 2025, n.62, suppl.1, e20515097.  Epub Dec 11, 2025. ISSN 0120-4688.  https://doi.org/10.25100/pfilosofica.v0i62s.15097.

This article proposes a philosophical and technical conceptualization of Large Language Models (LLMs), understood as statistical-probabilistic word calculators. The proposal is built upon a critical analogy with the mathematical calculator. The first section presents the structure and functioning of LLMs. First, we explore the limits of the comparison to highlight the potential of LLMs. Second, we explain the mathematical foundations of LLMs, specifically linear algebra. Finally, we outline their basic statistical operations through transformer architectures. The second section focuses on what is specific to LLMs as word calculators, distinguishing them from databases, analyzing the phenomenon of hallucination, and concluding with the possibilities that LLMs open for writing.

Keywords : LLM; Linear Algebra; Epistemology of AI; Neural Networks; Transformer architecture.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )